[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Sharing thoughts about the ddtp approaches

Hoi Jean,

On Sat, 5 Oct 2019 at 11:27, Jean-Philippe MENGUAL <jpmengual@debian.org> wrote:
> My concern is that we are very few contrib, and the work is giant. Many
> packages, but also many movements in packages (descriptions updats,
> additional packages, etc). I do not see how human beings can hope to
> translate all the packages descs in such conditions someday and for how
> long. Also because there as much work for translating as reviewing.

Thanks for the work. I agree that the backend of DDTP has a lot of
things that could be improved.

Some background: in the DDTP backend you are not translating
descriptions, but paragraphs; a description consists of multiple

This means that in the backend there are actually many little
optimisations that could be done. For example, if there is only a
small change between two paragraphs (for example a fixed typo, or
punctuation) then you could just copy the translations for those
paragraphs. I think the system does this already to some extent, I

One idea I had a long time ago was to automate a lot of the cases you
mention. Basically, if you see a paragraph like:

foo X bar

and the translations look like:

feu X bah

then if you see another paragraph that looks like

foo Y bar

then you can automatically create a translation like

feu Y bah

This should automatically handle a several of cases you suggest. X
could be package names, version numbers, etc.... Honestly, you could
probably do all sorts of fancy machine learning algorithms as well.

Many years ago grisu and I spent quite some time making a DDTP-ng,
which re-implemented DDTP/DDTSS as a Django project written in Python
with real database models where it would be easy to implement these
kinds of algorithms. The project unfortunately died, you should be
able to find it in the archives somewhere. One of the things it wanted
to support was for example having some translation engine provide
suggestions for translations of paragraphs and then an interface to
approve/adjust these suggestions. And a po interface for more other
possibly more advanced translation systems.

Maybe in the mean time more powerful translation systems have appeared
that DDTP could simply integrate with, and we could do away with the
current DDTSS altogether (it's terrible, I wrote it, I should know).

The main problem now is that the backend of DDTP/DDTSS is effectively
on life-support and has been for a very long time. I believe that to
really make the life of translators easier it would require work on
the order of DDTP-ng to bring the code up to a level other people want
to work with it. While translation teams are reasonably organised, the
backend simply doesn't get much love. I occasionally give it a kick
but that's it., I don't have time for much else.

Which reminds me, DDTP probably needs to start processing buster.

Not sure if this helps you. Any large change I think will require the
buy in of the translators. It totally not clear to me how important
this project actually is, but I feel it could be much better than it

Martijn van Oosterhout <kleptog@gmail.com> http://svana.org/kleptog/

Reply to: